Knowsis builds personas from the drivers of the outcome you care about, proves they hold on people the model has never seen, and puts them to work on the decisions that follow: which segment to back, which concept to take forward, what the brief should say.
Personas told apart people the model had never seen three times better than age, income and region did. Tested on public data, method fixed in writing first.
Four Money Manager personas now guide payment strategy across the Intrum European Consumer Payment Report.
The argument behind the method, set out in full in Greenbook, September 2026.
Trusted by
Clients and partners across Knowsis analytics work
Most insights teams already own a segmentation. It was expensive, it was accurate, and it describes the market well. Then someone asks it to make a decision, and it goes quiet. The problem is usually one of these.
Segments built on demographics and attitudes are easy to profile and hard to act on. Two people in the same segment make opposite choices, and the segmentation cannot say why, because it was never built on the drivers of the outcome you are trying to shift.
The launch workshop was good. Six months later the personas are a PDF nobody opens, and every new question about them means a new project. A persona that cannot be interrogated is a poster, not a tool.
A concept needs ranking, a message needs a first read, a price point needs a sanity check, and the next wave is three months away. The choice becomes gut feel or nothing.
Not a platform, not a framework. A persona set you can defend, and the means to use it.
We start from the decision, not the demographics. We model what actually moves the outcome you care about (loyalty, resilience, conversion, retention) for each respondent, then group people by the pattern of what moves them. Same survey data, organised around the thing you need to change.
Every persona set comes with three checks, run and reported: how well the personas hold on respondents held back from the build, how far apart the personas really are, and an audit of which answers are measured, which are modelled and which are inferred. If a check fails, you hear about it before the workshop, not after.
Each persona becomes something your team can interrogate in plain language: how would this group react to this concept, what would make them switch, what should the brief say. Every answer is labelled Measured, Modelled or Inferred, so nobody mistakes an inference for a finding.
The output is a ranking, a priority, a brief. Which segment to back. Which of five concepts to take to fieldwork. Where the next rand or euro of investment moves the number. That is what the personas were built to do.
We ran our persona method on a public dataset of 2,058 people, the one behind the study most often cited as evidence that AI personas fail. Ours separated people the model had never seen three times better than demographic segments, and held at full strength two weeks later. We also found out why honest people keep reaching opposite conclusions about this technology.
Read the validation →Twenty thousand consumers across twenty European markets, scored on a single financial health index and grouped into four Money Manager personas. The driver model ranked childhood money stress as roughly three times as important a predictor of adult financial fragility as current income. Payment strategy is now tailored by persona.
Read the case study → Download the full reportThe argument this method was built to test, published in Greenbook in September 2026.
Read it on Greenbook →
Six minutes. Four personas speak, every line labelled, and one of them declines to answer.
Watch on YouTube →Five steps. The third one is the whole argument.
The motivational layer. Where the questionnaire carries a short trade-off exercise, the Impulse Engine adds how each persona chooses under pressure, not only what it values at rest. The Impulse Engine →
You already have a segmentation study with respondent-level data. We rebuild it as predictive personas, run the validation, and hand back the persona set with the numbers. The fastest route, and the one most clients take first.
You have a decision and no study yet. We design the questionnaire around the outcome, including the short trade-off exercise the motivational layer needs, field it through your supplier or ours, and deliver the personas and the engine.
You have personas already and want them interrogable. We build the Persona Engine on top of your existing analysis, with the answer labelling and validation checks, and scope enablement so the team actually uses it.
What you need to bring: respondent-level survey data with an outcome worth predicting. For the motivational layer, about three minutes of questionnaire space. Enablement is part of the scope, not an afterthought, because a persona nobody talks to is back in the deck.
We publish the thinking behind the work as we go, including the tests that did not go the way we expected. Most of it is about what a persona or a digital twin can and cannot be trusted to do.
Three teams arrived at the same fix this month, from three directions. I ran it on the Columbia data and put a price on what the synthetic half is actually worth. About one extra respondent in seven.
You paid by the respondent. You bought one opinion, repeated two thousand times.
Last week Insight Innovation Ventures, in one of the most widely read newsletters in this industry, gave buyers some good advice.
Gold, in our business, is a decision you can defend. Bring the one you are stuck on and we will tell you in thirty minutes whether personas are the right tool for it, and what it would take.
You will speak to Greg Streatfield, founder of Knowsis, who has spent over twenty years building segmentations and driver models at TNS (now Kantar), Ogilvy and dunnhumby. About Greg
